Apply Machine Learning to the Interpretation of Urinary Crystal Morphology.

Kidney stones are primarily formed due to the supersaturation of ions in urine, leading to the formation of crystals. An assessment of the risk of kidney stones is based on a patient's medical history, biochemical urine tests, and various laboratory examinations. Combining these with imaging studies such as CT scans, ultrasound, and X-rays helps in diagnosing the type of kidney stones, though imaging results for smaller stones may be less accurate. Stone formation is common with a high recurrence rate, and there is a strong correlation between urine crystals and stone composition. Therefore, the analysis of urine crystals is meaningful for the diagnosis, evaluation of treatment strategies, and prevention of stone recurrence in kidney stone disease.

Microscopic analysis of urine crystals allows the observation of smaller crystals. However, manual urine microscopy is slow and time-consuming. To address this, we aim to develop artificial intelligence software to assist in the interpretation of urine crystals, providing a faster analysis. We will retrospectively analyze urine crystal images stored from previous research (Chang Gung Memorial Hospital Internal Project Research No. 107123-E) to identify crystal types. Subsequent image preprocessing and category labeling will be done to train and infer machine software. The results will be compared with manual interpretation to establish the accuracy of the software.